CoWeaver: A Bi-directional, Learnable and Explainable Matching Engine for Mixed Human-Agent Science Collaboration

2026-07-16

AI Summary

This work addresses the challenge of effectively coordinating large language model (LLM) agents with human scientists in collaborative scientific settings, where bidirectional dynamics and high interpretability requirements hinder seamless cooperation. To this end, the authors propose a novel bidirectional matching engine that integrates a two-stage ranking-and-filtering mechanism, uncertainty-aware capability modeling, and an upper confidence bound (UCB)-based exploration strategy. A feedback-driven mechanism dynamically updates estimates of agent and human capabilities over time. This approach uniquely unifies bidirectional matching, interpretability, and uncertainty-guided exploration within a hybrid human–AI scientific collaboration framework. Experimental results demonstrate that the method outperforms the purely greedy optimal solution on 6 out of 20 tasks while matching its performance on the rest, achieving significantly superior overall matching quality and efficiency compared to baseline approaches.

Abstract

LLM-based agents excel at writing articles, coding and information retrieval. However, they fail to form strong collaborations within the scientific community due to the bidirectional, dynamic nature of the problem and a high demand of decision interpretability. We proposed COWEAVER, a bidirectional, learnable and explainable algorithm to match scientists and form strong collaborations within a human-agent network. COWEAVER matches candidates and requesters through filling capability gaps and filters candidates through a two-stage ranking step. Finally, the model explores newcomers by maintaining uncertainty-aware capability estimates and updating them through requester's feedback. We show that the selection mechanism of combining both exploration (UCB) and greedy of COWEAVER exceeds the greedy-only mechanism - the analytical best solution - on 6 out of the 20 tasks and performed on par with the greedy-only mechanism in terms of selecting the best candidate. We compared COWEAVER baselines in terms of matching quality and efficiency. COWEAVER outperforms baselines on all metrics.

Problem

• human-agent collaboration
• scientific collaboration
• bidirectional matching
• explainable AI
• capability matching

Innovation

• bidirectional matching
• learnable collaboration
• explainable AI
• uncertainty-aware estimation
• human-agent teamwork